Multimodal branched transport infers anatomically aligned brain reaction maps

A new mathematical framework published on arXiv by researcher Cristian Mendico tackles a fundamental neuroscience problem: mapping how external stimulation propagates through the brain's architecture to create distributed reaction patterns. The paper, 'Multimodal branched transport infers anatomically aligned brain reaction maps,' introduces a variational optimization method that goes beyond existing large-scale control models. Instead of calculating transition costs on a prescribed network, Mendico's model infers the optimal routing map itself by combining three key data modalities: task-related blood-oxygen-level-dependent (BOLD) fMRI responses, source-reconstructed electrophysiology (EEG/MEG), and white-matter tractography to measure structural anisotropy. This fusion defines a biologically grounded 'anatomical transport cost.'

The core innovation is the application of 'branched transport' theory, a mathematical concept from optimal transport that favors the aggregation of flow into shared trunks before branching out to targets. Applied to neural signaling, this means the model identifies consolidated 'neural highways' where signals merge for efficient transport before diverging to their destinations. Mendico further attaches a stochastic graph-induced dynamics to the inferred map, allowing quantification of a crucial trade-off: geometric routing efficiency versus the dynamical controllability of the network. The results show that using multimodal, anatomically informed data generates more biologically plausible maps, that accounting for directional white-matter pathways (anisotropy) qualitatively reshapes the predicted routing backbones compared to isotropic models, and that optimizing for both geometry and dynamics reveals non-trivial performance reversals across different branching regimes.